Identifying subgroup markers in heterogeneous populations.

Traditional methods that aim to identify biomarkers that distinguish between two groups, like Significance Analysis of Microarrays or the t-test, perform optimally when such biomarkers show homogeneous behavior within each group and differential behavior between the groups. However, in many applications, this is not the case. Instead, a subgroup of samples in one group shows differential behavior with respect to all other samples. To successfully detect markers showing such imbalanced patterns of differential signal, a different approach is required. We propose a novel method, specifically designed for the Detection of Imbalanced Differential Signal (DIDS). We use an artificial dataset and a human breast cancer dataset to measure its performance and compare it with three traditional methods and four approaches that take imbalanced signal into account. Supported by extensive experimental results, we show that DIDS outperforms all other approaches in terms of power and positive predictive value. In a mouse breast cancer dataset, DIDS is the only approach that detects a functionally validated marker of chemotherapy resistance. DIDS can be applied to any continuous value data, including gene expression data, and in any context where imbalanced differential signal is manifested.

SERPINA6, BEX1, AGTR1, SLC26A3, and LAPTM4B are markers of resistance to neoadjuvant chemotherapy in HER2-negative breast cancer.

Response rates to chemotherapy remain highly variable in breast cancer patients. We set out to identify genes associated with chemotherapy resistance. We analyzed what is currently the largest single-institute set of gene expression profiles derived from breast cancers prior to a single neoadjuvant chemotherapy regimen (dose-dense doxorubicin and cyclophosphamide). We collected, gene expression-profiled, and analyzed 178 HER2-negative breast tumor biopsies ("NKI dataset"). We employed a recently developed approach for detecting imbalanced differential signal (DIDS) to identify markers of resistance to treatment. In contrast to traditional methods, DIDS is able to identify markers that show aberrant expression in only a small subgroup of the non-responder samples. We found a number of markers of resistance to anthracycline-based chemotherapy. We validated our findings in three external datasets, totaling 456 HER2-negative samples. Since these external sets included patients who received differing treatment regimens, the validated markers represent markers of general chemotherapy resistance. There was a highly significant overlap in the markers identified in the NKI dataset and the other three datasets. Five resistance markers, SERPINA6, BEX1, AGTR1, SLC26A3, and LAPTM4B, were identified in three of the four datasets (p value overlap < 1 × 10(-6)). These five genes identified resistant tumors that could not have been identified by merely taking ER status or proliferation into account. The identification of these genes might lead to a better understanding of the mechanisms involved in (clinically) observed chemotherapy resistance and could possibly assist in the recognition of breast cancers in which chemotherapy does not contribute to response or survival.

Platform comparisons for identification of breast cancers with a BRCA-like copy number profile.

Previously, we employed bacterial artificial chromosome (BAC) array comparative genomic hybridization (aCGH) profiles from BRCA1 and -2 mutation carriers and sporadic tumours to construct classifiers that identify tumour samples most likely to harbour BRCA1 and -2 mutations, designated 'BRCA1 and -2-like' tumours, respectively. The classifiers are used in clinical genetics to evaluate unclassified variants, and patients for which no good quality germline DNA is available. Furthermore, we have shown that breast cancer patients with BRCA-like tumour aCGH profiles benefit substantially from platinum-based chemotherapy, potentially due to their inability to repair DNA double strand breaks (DSB), providing a further important clinical application for the classifiers. The BAC array technology has been replaced with oligonucleotide arrays. To continue clinical use of existing classifiers, we mapped oligonucleotide aCGH data to the BAC domain, such that the oligonucleotide profiles can be employed as in the BAC classifier. We demonstrate that segmented profiles derived from oligonucleotide aCGH show high correlation with BAC aCGH profiles. Furthermore, we trained a support vector machine score to objectify aCGH profile quality. Using the mapped oligonucleotide aCGH data, we show equivalence in classification of biologically relevant cases between BAC and oligonucleotide data. Furthermore, the predicted benefit of DSB inducing chemotherapy due to a homologous recombination defect is retained. We conclude that oligonucleotide aCGH data can be mapped to and used in the previously developed and validated BAC aCGH classifiers. Our findings suggest that it is possible to map copy number data from any other technology in a similar way.

Selection of patients for hepatic surgery of colorectal cancer liver metastasis based on genomic aberrations.

BACKGROUND: We investigated whether genomic aberrations in primary colorectal cancer (CRC) can identify patients who are at increased risk of developing additional hepatic recurrence after colorectal liver metastases (CLM) resection. METHODS: Primary tumour DNA from 79 CLM resected patients was analysed for recurrent copy number changes (12x135k NimbleGen(™) aCGH). The cohort was divided into three groups: CLM patients with a recurrence-free survival after hepatic resection of at least 5 years (n = 21), patients who developed intra-hepatic recurrence (n = 32), and patients who developed extrahepatic recurrence (n = 26). By contrasting the primary tumour profiles of recurrence free and the extrahepatic recurrence CLM patients, a classifier, the extra-hepatic recurrence classifier (ERC1), predictive for subsequent extrahepatic-recurrence was developed. RESULTS: The ERC1 had an accuracy of 70 % (95 % confidence interval (CI): 55-82 %, misclassification error 30 %, base error rate: 45 %). This analysis identified a region on Chromosome 12p13 as differentially aberrated between these two groups. The classifier was further optimized by contrasting the extrahepatic recurrence group with the combined group of intrahepatic and no recurrence group, resulting in an extrahepatic prognostic classifier (ERC2) able to classify patients with CLMs suitable for hepatic resection with 74 % accuracy (95 % CI: 62-83 %, misclassification error 26 %, base error rate: 32 %). CONCLUSIONS: Patients with CLM who will develop extrahepatic recurrence may be identified with ERCs based on information in the primary tumour. Risk estimates for the occurrence of extrahepatic metastases may allow a reduction of hepatic resections of colorectal liver metastases for those who are unlikely to develop extrahepatic metastases.

Lobular histology and response to neoadjuvant chemotherapy in invasive breast cancer.

Invasive lobular carcinoma (ILC) has been reported to be less responsive to neoadjuvant chemotherapy (NAC) than invasive ductal carcinoma (IDC). We sought to determine whether ILC histology indeed predicts poor response to NAC by analyzing tumor characteristics such as protein expression, gene expression, and imaging features, and by comparing NAC response rates to those seen in IDC after adjustment for these factors. We combined datasets from two large prospective NAC trials, including in total 676 patients, of which 75 were of lobular histology. Eligible patients had tumors ≥3 cm in diameter or pathologic documentation of positive nodes, and underwent serial biopsies, expression microarray analysis, and MRI imaging. We compared pathologic complete response (pCR) rates and breast conservation surgery (BCS) rates between ILC and IDC, adjusted for clinicopathologic factors. On univariate analysis, ILCs were significantly less likely to have a pCR after NAC than IDCs (11 vs. 25 %, p = 0.01). However, the known differences in tumor characteristics between the two histologic types, including hormone receptor (HR) status, HER2 status, histological grade, and p53 expression, accounted for this difference with the lowest pCR rates among HR+/HER2- tumors in both ILC and IDC (7 and 5 %, respectively). ILC which were HR- and/or HER2+ had a pCR rate of 25 %. Expression subtyping, particularly the NKI 70-gene signature, was correlated with pCR, although the small numbers of ILC in each group precluded significant associations. BCS rate did not differ between IDC and ILC after adjusting for molecular characteristics. We conclude that ILC represents a heterogeneous group of tumors which are less responsive to NAC than IDC. However, this difference is explained by differences in molecular characteristics, particularly HR and HER2, and independent of lobular histology.

Concordance of clinical and molecular breast cancer subtyping in the context of preoperative chemotherapy response.

ER, PR and HER2 status in breast cancer are important markers for the selection of drug therapy. By immunohistochemistry (IHC), three major breast cancer subtypes can be distinguished: Triple negative (TN(IHC)), HER2+(IHC) and Luminal(IHC) (ER+(IHC)/HER2-(IHC)). By using the intrinsic gene set defined by Hu et al. five molecular subtypes (Basal(mRNA), HER2+(mRNA), Luminal A(mRNA), Luminal B(mRNA) and Normal-like(mRNA)) can be defined. We studied the concordance between analogous subtypes and their prediction of response to neoadjuvant chemotherapy. We classified 195 breast tumors by both IHC and mRNA expression analysis of patients who received neoadjuvant treatment at the Netherlands Cancer institute for Stage II-III breast cancer between 2000 and 2007. The pathological complete remission (pCR) rate was used to assess chemotherapy response. The IHC and molecular subtypes showed high concordance with the exception of the HER2+(IHC) group. 60% of the HER2+(IHC) tumors were not classified as HER2+(mRNA). The HER2+(IHC)/Luminal A or B(mRNA) group had a low response rate to a trastuzumab-chemotherapy combination with a pCR rate of 8%, while the HER2+(mRNA) group had a pCR rate of 54%. The Luminal A(mRNA) and Luminal B(mRNA) groups showed similar degrees of response to chemotherapy. Neither the PR status nor the endocrine responsiveness index subdivided the ER+(IHC) tumors accurately into Luminal A(mRNA) and Luminal B(mRNA) groups. Molecular subtyping suggests the existence of a HER2+(IHC)/Luminal(mRNA) group that responds poorly to trastuzumab-based chemotherapy. For Luminal(IHC) and triple negative(IHC) tumors, further subdivision into molecular subgroups does not offer a clear advantage in treatment selection.

Breast cancer subtype specific classifiers of response to neoadjuvant chemotherapy do not outperform classifiers trained on all subtypes.

INTRODUCTION: Despite continuous efforts, not a single predictor of breast cancer chemotherapy resistance has made it into the clinic yet. However, it has become clear in recent years that breast cancer is a collection of molecularly distinct diseases. With ever increasing amounts of breast cancer data becoming available, we set out to study if gene expression based predictors of chemotherapy resistance that are specific for breast cancer subtypes can improve upon the performance of generic predictors. METHODS: We trained predictors of resistance that were specific for a subtype and generic predictors that were not specific for a particular subtype, i.e. trained on all subtypes simultaneously. Through a rigorous double-loop cross-validation we compared the performance of these two types of predictors on the different subtypes on a large set of tumors all profiled on the same expression platform (n = 394). We evaluated predictors based on either mRNA gene expression or clinical features. RESULTS: For HER2+, ER- breast cancer, subtype specific predictor based on clinical features outperformed the generic, non-specific predictor. This can be explained by the fact that the generic predictor included HER2 and ER status, features that are predictive over the whole set, but not within this subtype. In all other scenarios the generic predictors outperformed the subtype specific predictors or showed equal performance. CONCLUSIONS: Since it depends on the specific context which type of predictor - subtype specific or generic- performed better, it is highly recommended to evaluate both specific and generic predictors when attempting to predict treatment response in breast cancer.

Gene expression profiles from formalin fixed paraffin embedded breast cancer tissue are largely comparable to fresh frozen matched tissue.

BACKGROUND AND METHODS: Formalin Fixed Paraffin Embedded (FFPE) samples represent a valuable resource for cancer research. However, the discovery and development of new cancer biomarkers often requires fresh frozen (FF) samples. Recently, the Whole Genome (WG) DASL (cDNA-mediated Annealing, Selection, extension and Ligation) assay was specifically developed to profile FFPE tissue. However, a thorough comparison of data generated from FFPE RNA and Fresh Frozen (FF) RNA using this platform is lacking. To this end we profiled, in duplicate, 20 FFPE tissues and 20 matched FF tissues and evaluated the concordance of the DASL results from FFPE and matched FF material. METHODOLOGY AND PRINCIPAL FINDINGS: We show that after proper normalization, all FFPE and FF pairs exhibit a high level of similarity (Pearson correlation >0.7), significantly larger than the similarity between non-paired samples. Interestingly, the probes showing the highest correlation had a higher percentage G/C content and were enriched for cell cycle genes. Predictions of gene expression signatures developed on frozen material (Intrinsic subtype, Genomic Grade Index, 70 gene signature) showed a high level of concordance between FFPE and FF matched pairs. Interestingly, predictions based on a 60 gene DASL list (best match with the 70 gene signature) showed very high concordance with the MammaPrint® results. CONCLUSIONS AND SIGNIFICANCE: We demonstrate that data generated from FFPE material with the DASL assay, if properly processed, are comparable to data extracted from the FF counterpart. Specifically, gene expression profiles for a known set of prognostic genes for a specific disease are highly comparable between two conditions. This opens up the possibility of using both FFPE and FF material in gene expressions analyses, leading to a vast increase in the potential resources available for cancer research.

KC-SMARTR: An R package for detection of statistically significant aberrations in multi-experiment aCGH data.

BACKGROUND: Most approaches used to find recurrent or differential DNA Copy Number Alterations (CNA) in array Comparative Genomic Hybridization (aCGH) data from groups of tumour samples depend on the discretization of the aCGH data to gain, loss or no-change states. This causes loss of valuable biological information in tumour samples, which are frequently heterogeneous. We have previously developed an algorithm, KC-SMART, that bases its estimate of the magnitude of the CNA at a given genomic location on kernel convolution (Klijn et al., 2008). This accounts for the intensity of the probe signal, its local genomic environment and the signal distribution across multiple samples. RESULTS: Here we extend the approach to allow comparative analyses of two groups of samples and introduce the R implementation of these two approaches. The comparative module allows for a supervised analysis to be performed, to enable the identification of regions that are differentially aberrated between two user-defined classes.We analyzed data from a series of B- and T-cell lymphomas and were able to retrieve all positive control regions (VDJ regions) in addition to a number of new regions. A t-test employing segmented data, that we implemented, was also able to locate all the positive control regions and a number of new regions but these regions were highly fragmented. CONCLUSIONS: KC-SMARTR offers recurrent CNA and class specific CNA detection, at different genomic scales, in a single package without the need for additional segmentation. It is memory efficient and runs on a wide range of machines. Most importantly, it does not rely on data discretization and therefore maximally exploits the biological information in the aCGH data.The program is freely available from the Bioconductor website http://www.bioconductor.org/ under the terms of the GNU General Public License.